Legacy precoding matrix index (PMI) selection methods based on codebook search are increasingly impractical due to the large codebook size. Existing implementation usually use greedy search methods to select the PMI which has high complexity, resulting in low performance.